Water　environment　time　series　prediction　is　important　to　efficient　water　resource　manage-ment.　Traditional　water　quality　prediction　is　mainly　based　on　linear　models.　However,　owing　to　complex　conditions　of　the　water　environment,　there　is　a　lot　of　noise　in　the　water　quality　time　series,　which　will　seriously　affect　the　accuracy　of　water　quality　prediction.　In　addition,　linear　models　are　difficult　to　deal　with　the　nonlinear　relations　of　data　of　time　ser-ies.　To　address　this　challenge,　this　work　proposes　a　hybrid　model　based　on　a　long　short　-term　memory-based　encoder-decoder　neural　network　and　a　Savitzky-Golay　filter.　Among　them,　the　filter　of　Savitzky-Golay　can　eliminate　the　potential　noise　in　the　time　series　of　water　quality,　and　the　long　short-term　memory　can　investigate　nonlinear　characteristics　in　a　complicated　water　environment.　In　this　way,　an　integrated　model　is　proposed　and　effectively　obtains　statistical　characteristics.　Realistic　data-based　experiments　prove　that　its　prediction　performance　is　better　than　its　several　state-of-the-art　peers.　(c)　2021　Elsevier　Inc.　All　rights　reserved.